[compiler-rt] Replace direct calls to pipe with internal_pipe

[Lancern]

This PR tries to resolve google/sanitizers#1106 by introducing a new internal_pipe function which will not be intercepted by TSAN.

[llvmbot]

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Sirui Mu (Lancern)

Changes

This PR tries to resolve google/sanitizers#1106 by introducing a new internal_pipe function which will not be intercepted by TSAN.

---

Full diff: https://github.com/llvm/llvm-project/pull/97186.diff

5 Files Affected:

• (modified) compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp (+4)
• (modified) compiler-rt/lib/sanitizer_common/sanitizer_mac.cpp (+4)
• (modified) compiler-rt/lib/sanitizer_common/sanitizer_netbsd.cpp (+5)
• (modified) compiler-rt/lib/sanitizer_common/sanitizer_posix.h (+2)
• (modified) compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp (+1-1)

diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp
index 12df3ef73da4b..c70a689c7e599 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_linux.cpp
@@ -796,6 +796,10 @@ uptr internal_lseek(fd_t fd, OFF_T offset, int whence) {
   return internal_syscall(SYSCALL(lseek), fd, offset, whence);
 }
 
+uptr internal_pipe(fd_t pipefd[2]) {
+  return internal_syscall(SYSCALL(pipe), pipefd);
+}
+
 #    if SANITIZER_LINUX
 uptr internal_prctl(int option, uptr arg2, uptr arg3, uptr arg4, uptr arg5) {
   return internal_syscall(SYSCALL(prctl), option, arg2, arg3, arg4, arg5);
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_mac.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_mac.cpp
index cbdf3e95925bf..f11d681f2a388 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_mac.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_mac.cpp
@@ -213,6 +213,10 @@ uptr internal_unlink(const char *path) {
   return unlink(path);
 }
 
+uptr internal_pipe(fd_t fildes[2]) {
+  return pipe(fildes);
+}
+
 uptr internal_sched_yield() {
   return sched_yield();
 }
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_netbsd.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_netbsd.cpp
index 5e601bdcde1e5..eff8ceb3b60b9 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_netbsd.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_netbsd.cpp
@@ -204,6 +204,11 @@ uptr internal_rename(const char *oldpath, const char *newpath) {
   return _REAL(rename, oldpath, newpath);
 }
 
+uptr internal_pipe(fd_t fildes[2]) {
+  DEFINE__REAL(int, pipe, int a[2]);
+  return _REAL(pipe, fildes);
+}
+
 uptr internal_sched_yield() {
   CHECK(&_sys_sched_yield);
   return _sys_sched_yield();
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_posix.h b/compiler-rt/lib/sanitizer_common/sanitizer_posix.h
index 14617e4771bec..c0a93eabda57d 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_posix.h
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_posix.h
@@ -56,6 +56,8 @@ uptr internal_unlink(const char *path);
 uptr internal_rename(const char *oldpath, const char *newpath);
 uptr internal_lseek(fd_t fd, OFF_T offset, int whence);
 
+uptr internal_pipe(fd_t pipefd[2]);
+
 #if SANITIZER_NETBSD
 uptr internal_ptrace(int request, int pid, void *addr, int data);
 #else
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp b/compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp
index ece2d7d63dd61..93c49fa1dafa4 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_posix_libcdep.cpp
@@ -293,7 +293,7 @@ bool IsAccessibleMemoryRange(uptr beg, uptr size) {
   // Checking too large memory ranges is slow.
   CHECK_LT(size, page_size * 10);
   int sock_pair[2];
-  if (pipe(sock_pair))
+  if (internal_pipe(sock_pair))
     return false;
   uptr bytes_written =
       internal_write(sock_pair[1], reinterpret_cast<void *>(beg), size);

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[cjappl]

This seems reasonable to me.

Please:

• Fix the code formatting issue
• Add a test that catches this issue and proves it doesn't happen anymore

I'm going to add one of the original commenters on your issue to this PR.

[Lancern]

@cjappl Sorry for the delay here. Updated this PR and added a test that verify TSAN does not complain any more.